An automated technique for adaptive radar polarimetric pattern classification is
described. The approach is based on a genetic algorithm that uses a probabilistic pattern
separation distance function and searches for those transmit and receive states of polarization
sensing angles that optimize this function. Seven pattern separation distance functions—the
Rayleigh quotient, the Bhattacharyya, divergence, Kolmogorov, Matusta, Kullback–Leibler
distances, and the Bayesian probability of error—are used on real, fully polarimetric synthetic
aperture radar target signatures. Each of these signatures is represented as functions of transmit and
receive polarization ellipticity angles and the angle of polarization ellipse. The results
indicate that, based on the majority of the distance functions used, there is a unique set of state of
polarization angles whose use will lead to improved classification performance.
© 2006 Optical Society of America
Equations on this page are rendered with MathJax. Learn more.